Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2025 Mar 1.
Published in final edited form as: Biol Psychiatry Cogn Neurosci Neuroimaging. 2023 Sep 9;9(3):326–336. doi: 10.1016/j.bpsc.2023.08.010

Machine Learning of Functional Connectivity to Biotype Alcohol and Nicotine Use Disorders

Tan Zhu 1, Wuyi Wang 2, Yu Chen 3, Henry R Kranzler 4, Chiang-Shan R Li 3,5,6, Jinbo Bi 1,*
PMCID: PMC10976073  NIHMSID: NIHMS1947421  PMID: 37696489

Abstract

Background

Magnetic resonance imaging provides noninvasive tools to investigate alcohol and nicotine use disorders (AUD/NUD) and neural phenotypes for genetic studies. A data-driven transdiagnostic approach could provide a new perspective on the neurobiology of AUD and NUD.

Methods

Using samples of individuals with AUD (N=140), NUD (N=249), and healthy controls (HCs) (N=461) from the UK Biobank, we integrated clinical, neuroimaging, and genetic markers to identify biotypes of AUD and NUD. We partitioned AUD and NUD subjects based on resting-state functional connectivity features (FCs) associated with clinical metrics. A multi-task artificial neural network was trained to evaluate the cluster-defined biotypes and jointly infer AUD and NUD diagnoses.

Results

Three biotypes – primary NUD, mixed NUD/AUD with depression and anxiety, and mixed AUD/NUD – were identified. Multi-task classifiers incorporating biotype knowledge achieved higher area under the curve (AUD: 0.76, NUD: 0.74) than single-task classifiers without biotype differentiation (AUD: 0.61, NUD: 0.64). Cerebellar FCs were important in distinguishing the three biotypes. The biotype of mixed NUD/AUD with depression and anxiety demonstrated the largest number of FCs (N=5), all related to visual cortex, that significantly differed from HCs and were validated in a replication sample (P<0.05). A polymorphism in TNRC6A was associated with the mixed AUD/NUD biotype in both the discovery (P=7.3E-05) and replication sets (P=4.2E-02).

Conclusions

Biotyping and multi-task learning using FCs can characterize the clinical and genetic profiles of AUD and NUD, and help identify cerebellar and visual circuit markers to differentiate AUD/NUD from HCs. These markers support a new, growing body of literature.

Keywords: Resting-state functional MRI, biotyping, machine learning, alcohol use disorder, nicotine use disorder, genetic association analysis

1. INTRODUCTION

Alcohol and nicotine use disorders (AUD and NUD, respectively) are etiologically heterogeneous and include sub-phenotypes with distinct profiles of use behaviors, consequences, and neurobiological underpinnings (1,2). Typological research has helped to refine the classification of substance use disorders in the Diagnostic and Statistical Manual of Mental Disorders and International Classification of Diseases (3,4). However, empirical subtyping of clinical characteristics does not reflect this heterogeneity or inform treatment decisions (5,6). Hence, to enhance the validity of subtyping, the National Institute of Mental Health developed research domain criteria that include biological and neural markers to identify “biotypes” (7–10). Accumulating evidence supports a transdiagnostic approach that cuts across traditional diagnostic boundaries or sets them aside to provide new insights into the etiology and manifestations of AUD and NUD. As smoking and drinking frequently co-occur (11), the neural circuit dysfunctions are unlikely to be pathognomonic for AUD or NUD in isolation (12–14). Innovative machine learning approaches can help to characterize the neural circuit distinct to or shared by AUD and NUD. Whereas genetic factors contribute to diagnostic heterogeneity in mental illnesses (6,15), integrating neuroimaging and genomic data could help in characterizing the heterogeneity and uncovering intermediate phenotypes for subgrouping individuals in a broad diagnostic entity (16).

Studies have successfully employed magnetic resonance imaging (MRI) to quantify structural and functional brain differences between individuals with AUD or NUD and healthy controls (17–20). For instance, alcohol misuse disproportionately affects the thalamus and hippocampus (21–24). Chronic, heavy drinking leads to thalamic dysfunction, which in the extreme manifests as Korsakoff’s syndrome (25). Smoking can impact brain function via the basal forebrain circuits, which, as demonstrated in preclinical studies, support attentional and cognitive control (26). Imaging studies have also demonstrated altered functional connectivity (FC) of the brain in association with alcohol and nicotine misuse (27–30), implicating cortico-striatal dysconnectivity during resting state and task challenges in AUD (31–36) and NUD (37–39). However, it remains unclear whether the FC changes are specific to or shared by frequently co-occurred AUD and NUD.

Here, we utilized UK Biobank (UKB) (40) data to identify neural biotypes of individuals with AUD and NUD based on their whole-brain FC features derived from resting-state fMRI (rsfMRI). First, we employed connectivity features to identify biotypes, which provided more accurate classifiers in distinguishing AUD and NUD from healthy controls (HCs) than a classification scheme using the same subjects without biotype differentiation. Second, we proposed and validated a novel artificial neural network (ANN) to leverage the biotype knowledge and jointly infer AUD and NUD via multi-task learning (MTL). MTL is a machine learning principle that estimates models for multiple inferential tasks jointly when the tasks (here, predicting AUD and NUD) are assumed to be related. Third, using the proposed approach, we identified FC markers that differentiate the biotypes. Finally, we explored genetic associations of the biotype to identify variants not evident by merely comparing AUD/NUD and HC.

2. METHOD AND MATERIALS

2.1. Subjects and clinical measures

We evaluated ICD-10 (41) diagnotic codes for 21,400 subjects with both rsfMRI and T1 anatomical imaging data in the UKB database, and categorized subjects with any subcode(s) of “F10” and “F17” as AUD and NUD subjects, respectively. We excluded from the AUD/NUD sample subjects with ICD-10 codes of other mental or any neurological disorders, except mild or unspecified depression and anxiety, and identified subjects without any ICD-10 code as HCs. The sample comprised 198 AUD, 499 NUD, and 4956 HC subjects. Additional UKB clinical variables used in our analysis include depressive and anxiety symptom (“recent feelings of depression” and “recent feelings of nervousness or anxiety” coded 1 : not at all; 2 : several days; 3 : more than half the days; and 4 : nearly every day to reflect increasing symptom severity) and ICD-10 codes for a personal history of psychoactive substance abuse (Z86.4), depression (F32.0 mild; F32.9 unspecified), and anxiety (F40.9 unspecified phobic; F41.9 unspecified anxiety).

We selected the largest groups of NUD and HC subjects that were matched to the AUD subjects (the smallest group) on age, gender, and race/ethnicity, and levels of anxiety and depressive symptoms. We excluded subjects with incomplete or poor-quality images, including those with head motion ≥1.5 mm during the rsfMRI scan. The final sample of 850 individuals comprised 140 AUD, 249 NUD, and 461 HCs (see Supplementary Table S2 for statistics).

2.2. Functional connectivity features

UKB imaging data were acquired on a Siemens Skyra 3-Tesla MR scanner with a 32-channel head coil using protocols developed by the UKB Imaging Working Group. The imaging parameters for rsfMRI have been described previously (42,43). Briefly, the resting-state data were collected with eyes open for 6 minutes (TR = 0.735 s, with multiband acceleration factor = 8).

The rsfMRI data were processed with SPM12 (Statistical Parametric Mapping) (44). Each participant’s image was first realigned (motion corrected) and a mean functional image volume was constructed from the realigned image volumes. These mean images were co-registered with the high-resolution structural MPRAGE image and segmented for normalization with affine registration followed by nonlinear transformation. The normalization parameters determined for the structural volume were then applied to the corresponding functional image volumes for each participant. The images were then smoothed with a Gaussian kernel of 4 mm at Full Width at Half Maximum. Covariates of no interest were regressed out from the data, including linear, quadratic, and cubic drift, a 24-parameter model of motion parameters (45), as well as the mean cerebral-spinal-fluid, mean white matter, and global signals. Data were temporally smoothed with a zero-mean unit-variance low-pass Gaussian filter (cutoff frequency ~ 0.12 Hz). A correlation matrix of 268 regions of interest (ROIs) (46) yielded 35,778 unique FC markers for each individual subject.

2.3. Genotypes

Genotype data were available for 136 AUDs, 238 NUDs, and 452 HCs. DNA samples were extracted from blood and genotyped with the Applied Biosystems™ UK BiLEVE Axiom™ Array or Applied Biosystems™ UK Biobank Axiom™ Array (47). Imputed genotypes used the Haplotype Reference Consortium as the imputation reference panel. Although genome-wide single nucleotide polymorphisms (SNPs) were available, the sample size was not large enough to conduct a genome-wide association study (GWAS). Thus, we performed candidate gene analysis by identifying genes in the NHGRI-EBI GWAS Catalog (48) that were previously associated with AUD and NUD, and 12,772 SNPs in 84 AUD- and NUD-related genes available in the UKB.

2.4. Machine learning analysis

2.4.1. Overall analytic goals and routines

Of the 850 subjects, 765 (90%, with 126 AUDs, 225 NUDs, and 414 HCs) were used for discovery and 85 (14 AUDs, 24 NUDs, and 47 HCs) for replication. The discovery set was further split into training (n=680, including 111 AUDs, 201 NUDs, and 368 HCs) and validation (n=85, 15 AUDs, 24 NUDs, and 46 HCs) sets to allow validation of the classifiers without involving the replication set. Supplementary Table S3 shows the demographic and clinical characteristics of the training and replication samples and the statistics.

Figure 1 shows the analytic procedures. Although sex is a critical factor, because sex-stratified analyses reduce the available statistical power, we accounted for covariate effects (49,50) in the computation of each FC feature by regressing out age and sex in a linear model of the training data. Validation and replication datasets were corrected using the same linear models of training data. The training sample was used to identify biotypes with selected FC features and to create classifiers with the biotypes to differentiate AUD or NUD from HC. The validation sample was used to compare classifiers and select the model with the highest classification accuracy. The selected classifiers produced quantitative scores for each subject reflecting their membership in a specific biotype, which also served as quantitative traits in genetic association tests. During replication, the validated classifiers were applied to the replication sample to assess classification accuracy. FC and genetic markers identified during discovery were also tested for replication.

Figure 1.

Figure 1.

Open in a new tab

The analytic pipeline to identify and validate AUD/NUD biotypes and the FC and genetic markers differentiating the biotypes. MTL: multi-task learning

2.4.2. Feature selection, dimension reduction, and cluster analysis for biotype identification

First, we searched for top FC features that were significantly Spearman’s rank-correlated with any of the clinical features described in Section 2.1. To identify biotypes based on FC features related to AUD/NUD, we selected the top FC features that differed significantly between HCs and either AUDs or NUDs based on Wilcoxon rank-sum tests. We combined these features for cluster analysis, setting the significance threshold at P = 0.05, 0.01, 0.005 and 0.001 to identify four sets of FC features. We selected the set with the best biotyping performance.

We reduced the data dimension by the Uniform Manifold Approximation and Projection (UMAP), a novel nonlinear dimension reduction and visualization technique (51). UMAP preserves subject proximity in the sample when searching for a low dimensional space to map high dimensional data. We used UMAP to compress each of the four identified sets of FC features into 2, 3, and 4 dimensions, which resulted in 12 distinct UMAP representations. For each representation, K-means and hierarchical clustering were used to derive subject clusters, with the number of clusters (2, 3 and 4) determined by the mean Silhouette Coefficient, a ratio of the similarity within a cluster versus that of other clusters, as widely used in machine learning (52), bioinformatics (53) and neuroscience (54) research. We trained the classifiers for each of the 12 biotyping solutions and selected the best-performing solution. The final clusters were compared on demographic, clinical characteristics and FC features.

2.4.3. Differentiating AUD and NUD individuals from controls

Biotyping helped to reduce between-subject heterogeneity in alcohol- and nicotine-related FC features and facilitated the classification of AUD and NUD. We constructed an ANN that performed MTL to differentiate AUD or NUD from HC simultaneously. Biotype scores were calculated and then used to predict AUD or NUD subjects (workflow shown in Supplementary Figure S1). The resultant classifier was then applied to classify replication subjects for performance evaluation. The input to the ANN is a whole-brain FC matrix containing 35,778 distinct pair-wise correlations of the 268 ROIs. To reduce the number of trainable parameters in the ANN and the risk of overfitting the model, the ANN performed feature extraction to map the high-dimensional whole-brain FC matrix into a low-dimensional representation (Supplementary Methods).

To examine whether biotyping and/or MTL improved the classification, three ANN models were trained and their performance compared: (i) a model trained to classify AUD/NUD separately from HCs directly; (ii) a model trained to classify AUD/NUD jointly; and (iii) our approach, in which we first computed the biotype scores according to the low-dimensional representation as described above. We then leveraged MTL to jointly infer AUD and NUD scores, which allowed the classifiers to be trained using both AUD and NUD samples. To train this MTL model, we minimized a new objective function measuring biotype classification error together with the overall AUD and NUD (vs. control) classification error. Objective functions of the other two models are described in Supplementary Methods. By comparing the area under the receiver operating characteristic curves (AUCs) of the resultant AUD and NUD classifiers on the validation sample, we selected the best clustering solution and corresponding classifier.

Statistical significance was estimated by permutation-testing whether any of the above three classifiers performed significantly better than chance. We randomly permuted the AUD, NUD and HC labels in the discovery and replication sets. For each permuted discovery dataset, we trained ANN classifiers using the three procedures, classified AUD/NUD in the replication sample, and tested whether the permuted classification performance was significantly worse than the non-permuted ANNs. We repeated this procedure 1000 times and report the statistical significance of classification accuracy. Using bootstrapping (55), we examined whether biotype knowledge and MTL boost classification performance. For each of the three models, we sampled from the training, validation, and replication sets with replacement so that the three new (bootstrapped) sets were the same size as the original datasets. We used the bootstrapped sets to train, select, and evaluate the ANN model. After repeating the bootstrapping 1000 times, we report the statistical significance of the differences between the bootstrapping results for the three models.

2.4.4. Cross-validation of biotypes

We performed cross-validation by splitting all subjects in the training set into four validation folds of 170 each. For each fold, we trained our ANN model using the samples from the remaining three folds (N=510). With the fitted ANN model, we generated biotype predictions for subjects in the validation fold and calculated the AUC. This process was repeated for each of the four validation folds. The biotype validity was assessed by performing permutation tests (1000 trials) to examine whether the cross-validation AUCs were significantly better than chance (0.5). The validation folds were also used to determine the threshold for converting the probabilistic outputs of the final classifier into biotype assignments by minimizing the absolute distance between the specificity and sensitivity of the classifier. We applied this threshold to assign the 38 AUD/NUD subjects in the replication set to different biotypes to validate all FC features identified during discovery.

2.4.5. Genetic association analysis

The ANN classifier computed subjects’ membership scores for the biotypes at the biotype scoring layer and the scores measuring the probability of AUD and NUD at the final layer (Supplementary Figure S1). The biotype membership and AUD/NUD scores from the best ANN model were treated as quantitative traits in genetic association analyses to identify SNPs. We performed association tests using the Genome-wide Efficient Mixed-Model analysis for Association (GEMMA) (56) with the five traits using the training samples, with age and sex as covariates, and replicated the identified genetic markers on the replication set. GEMMA adopts a linear mixture model that corrects for correlations among related individuals by a genetic relationship matrix calculated from high-quality imputed genome-wide SNPs (info score ≥ 0.75) for all subjects. SNPs that were available for < 95% of the subjects or for which the Hardy–Weinberg equilibrium P-value was < 1.0E-07 were excluded. SNPs with minor allele frequency (MAF) < 5% were removed from the association tests for each trait. Finally, 3,096 SNPs from 80 genes remained (details in Supplementary Table S4). As Bonferroni correction for all SNPs would overcorrect for SNPs that are not truly independent, we estimated the number of independent SNPs (57). We identified blocks of linkage disequilibrium (LD) for the selected SNPs and evaluated the number by counting one SNP per LD block, plus all SNPs outside the blocks. Because >90% of the training and replication subjects were British White (BW), we examined the results from only BWs. The demographics of the samples are summarized in Supplementary Table S5. Results on the full sample were included in Supplementary Table S6 and S7.

3. RESULTS

3.1. AUD/NUD biotypes and the FC features

After correcting for age and sex, Spearman’s rank correlation and Wilcoxon rank-sum test based on the training set identified 9,344; 2,240; 1,155; and 237 FC features, respectively, for P < 0.05, 0.01, 0.005, and 0.001. For each of the four feature sets, UMAP reduced the data of 322 AUD and NUD subjects into lower dimensions (D=2, 3, and 4). Following the analytic pipeline described in Section 2.4, a 3-cluster biotyping solution obtained by K-means clustering of 237 FC features (corresponding to P < 0.001) in the 4D UMAP space showed the highest geometric mean of validation AUCs (AUD: 0.81, P = 0.003, NUD: 0.87, P < 0.001, permutation test) for the AUD and NUD classifier. Cross-validation of the three biotypes yielded AUC values of 0.77, 0.79, and 0.70, respectively, for biotypes 1, 2, and 3 (P < 0.005, permutation test, Supplementary Figure S2).

The biotypes and HCs were comparable demographically (Supplementary Table S8). Table 1 lists the clinical metrics that differed significantly among the three biotypes and between biotypes and HCs. All biotypes had higher scores than the HCs for most alcohol use and psychiatric metrics, differing significantly on the severity of depression and anxiety and the percentage of subjects with AUD or NUD. Whereas 81.4% of the subjects in biotype 1 had NUD, we named it “Primary NUD”. Biotypes 2 and 3 had mixed proportions of AUD and NUD with less AUD in biotype 2 than 3 (38.9% vs. 52.7%). Biotype 2 had more severe depressive and anxiety symptoms than biotypes 1 and 3. Thus, we named biotype 2 “mixed NUD/AUD with depression and anxiety (DE/AX)” and biotype 3 as “mixed AUD/NUD”.

Table 1.

Clinical characteristics of the three biotypes and controls.

Primary NUD (N=113) Mixed NUD/AUD with DE/AX (N=108) Mixed AUD/NUD (N=91) Control (N=368) P-value
Depression (online questionnaire)1 1.2(0.4) 1.7(0.9) 1.1(0.5) 1.4(0.7) *** Mixed NUD/AUD with DE/AX > Primary NUD, Mixed AUD/NUD, Control
* Control > Primary NUD, Mixed AUD/NUD
Anxiety (online questionnaire)1 1.2(0.5) 1.9(1.0) 1.3(0.5) 1.5(0.8) *** Mixed NUD/AUD with DE/AX > Primary NUD Mixed AUD/NUD, Control
* Control > Primary NUD
Depression (ICD-10), N(%)2 9(8.0) 19(17.6) 2(2.2) 0(0.0) *** Mixed NUD/AUD with DE/AX > Primary NUD, Mixed AUD/NUD, Control
*** Primary NUD, Mixed AUD/NUD > Control
Anxiety (ICD-10), N(%)2 4(3.5) 10(9.3) 0(0.0) 0(0.0) *** Mixed NUD/AUD with DE/AX > Mixed AUD/NUD, Control
* Mixed NUD/AUD with DE/AX > Primary NUD
Z86.4-Substance use, N(%)2 19(16.8) 25(23.1) 14(15.4) 0(0.0) *** Primary NUD, Mixed NUD/AUD with DE/AX, Mixed AUD/NUD > Control
NUD, N(%)2 92(81.4) 66(61.1) 43(47.3) 0(0.0) *** Primary NUD > Mixed NUD/AUD with DE/AX, Mixed AUD/NUD, Control
*** Mixed NUD/AUD with DE/AX > Mixed AUD/NUD, Control
*** Mixed AUD/NUD > Control
AUD, N(%)2 21(18.6) 42(38.9) 48(52.7) 0(0.0) *** Mixed AUD/NUD > Primary NUD and Control
** Mixed AUD/NUD > Mixed NUD/AUD with DE/AX
*** Mixed NUD/AUD with DE/AX > Primary NUD, Control
*** Primary NUD > Control
Open in a new tab

Note: For Anxiety and Depression (online questionnaire) where ratings of 1 to 4 were given, the mean and standard deviation (in parentheses) are reported. For other clinical metrics, the number of subjects with the corresponding ICD-10 codes and percentage of such subjects in each biotype (in parentheses) are reported.

AUD: alcohol use disorder, NUD: nicotine use disorder, DE: depression AX: anxiety.

1

Kruskal-Wallis H test was used to examine the differences of the three biotypes and control group. If significant differences were found (P < 0.05), Tukey post-hoc tests were conducted pairwise of the groups.

2

Discrete variables were tested by χ2 test. If significant differences were found among the three biotypes and control group (P < 0.05), rank sum tests were conducted pairwise of the groups.

***

P < 0.005,

**

P < 0.01,

*

P < 0.05, Bonferroni corrected.

We compared the biotypes and controls on the 237 selected FC features. Supplementary Table S9 provides the test results and reproducibility of the Wilcoxon rank-sum tests of the FC features that significantly differentiated any biotype from controls (P < 0.05, Bonferroni corrected, see Table S10 for more details). In Figure 2, with BrainNet Viewer (58), a visualization tool for human brain connectomics, we show the identified FC features that were validated in the replication set (P < 0.05). In Figure 3, we also distinguished the FC features amongst the three biotypes.

Figure 2.

Figure 2.

Open in a new tab

FC features significantly differentiating each of the biotypes: (A) Primary NUD, (B) Mixed NUD/AUD with depression and anxiety biotype, and (C) Mixed AUD/NUD, from HCs in both discovery and replication sets (Wilcoxon rank-sum tests, P < 0.05, Bonferroni corrected). ROIs were color-coded, with node size scaled by its node degree. The edges were color-coded according to the feature effects: red: FC values of the biotype > control, on average; blue: biotype < control, on average, while the edge thickness corresponded to the absolute value of the mean of FC feature difference. The number in the axial view of the brain is the index of ROI, whose axial, coronal, sagittal slice view, and the 3D rendered view can be found in Supplementary Figure S3. Detailed information about the FC features can be found in Supplementary Table S9a. In each panel, shown from left to right are the sagittal, axial, and coronal views of the brain. PSC: primary sensory cortex; VAC: visual association cortex; SVC; secondary visual cortex; IFG: inferior frontal gyrus (pars orbitalis); SMG: supramarginal gyrus; ITG: inferior temporal gyrus.

Figure 3.

Figure 3.

Open in a new tab

FC features significantly differentiating the three biotypes: (A) Primary NUD vs. Mixed NUD/AUD with depression and anxiety, (B) Primary NUD vs. Mixed AUD/NUD, and (C) Mixed NUD/AUD with depression and anxiety vs. Mixed AUD/NUD, in both discovery and replication (Wilcoxon rank-sum tests, P < 0.05, Bonferroni corrected). ROIs were color-coded and node size was scaled by its node degree. The edges were color-coded according to the feature effects: red: FC values of the former > latter biotype, on average; blue: former < latter biotype, on average, while the edge thickness corresponded to the absolute value of the mean of FC feature difference. The number in the axial view of the brain is the index of ROI, whose axial, coronal, sagittal slices, and the 3D rendered view can be found in Supplementary Figure S3. Detailed information about the FC features can be found in Supplementary Table S9b. In each panel, shown from left to right are the sagittal, axial, and coronal views of the brain. PSC: primary sensory cortex; SVC; secondary visual cortex; STG: Superior Temporal Gyrus.

3.2. Distinguishing AUD and NUD from HC: the effects of biotyping

We compared the AUCs of the three models in the replication set: baseline ANN, MTL ANN, and MTL ANN with biotype knowledge. With both biotyping and MTL, the ANN model achieved the highest AUC in predicting both AUD (0.76, P < 0.001, permutation test) and NUD (0.74, P = 0.001). Even ANN with only MTL (AUD: 0.75, P = 0.001, NUD: 0.68, P = 0.01) is better than the baseline ANN (no MTL, no biotyping, AUD: 0.61, P = 0.01, NUD: 0.64, P = 0.02). Bootstrapping tests showed that, in differentiating AUD and HC, MTL ANN with biotype knowledge (AUC = 0.71 ± 0.10) was significantly better than baseline ANN (AUC = 0.68 ± 0.09, P = 0.002) and MTL ANN (AUC = 0.68 ± 0.09, P = 0.01); the same finding was obtained for classifying NUD, where bootstrapping tests yielded P = 7.2E-6 (0.70 ± 0.08 vs. 0.67 ± 0.08) and P = 8.8E-4 (0.70 ± 0.08 vs. 0.68 ± 0.08), respectively (Figure 4).

Figure 4.

Figure 4.

Open in a new tab

Bootstrapping test (N=1000) for artificial neural network (ANN) classifiers: 1) ANN: baseline ANN separately trained to directly classify alcohol use disorder (AUD) or nicotine use disorder (NUD) from healthy controls (HCs); 2) ANN with multi task framework (MT-ANN): ANN trained to classify AUD and NUD at the same time but without biotypes knowledge; 3) ANN with multi task framework and biotype knowledge (MT-ANN-WB): ANN trained to classify AUD and NUD at the same time, and with the knowledge of AUD/NUD biotypes. * P < 0.05, ** P < 0.005, *** P < 0.0005.

3.3. Genetic association analysis

Among the 3,096 SNPs in the training set, we identified 364 LD blocks and 317 SNPs outside the blocks. Given 681 effective tests for Bonferroni correction, the significance threshold was set to 7.4E-05 in discovery and replication threshold was 0.05. Table 2 lists the top three SNPs identified for each of the three biotypes. No SNPs were significantly associated with an AUD or NUD diagnosis in discovery. One SNP (rs11646987) in TNRC6A was significantly associated with mixed AUD/NUD (P = 7.3E-05 in discovery and P = 0.042 in replication). With the full sample, the same SNP was identified with lower p-values (P = 4.9E-05 vs. P = 0.042).

Table 2.

Top single nucleotide polymorphisms (SNPs) with main effect on different traits, ranked by P-values.

Rank SNP Chr Gene MAF Beta Discovery P-value 1 Replication P-value 1
Primary NUD BIOTYPE
1 rs9933173 16 ZFHX3 0.068 1.951 3.9E-04 8.8E-01
2 rs2538414 7 DPP6 0.34 −0.981 7.9E-04 6.3E-01
3 rs12927436 16 ZFHX3 0.057 −1.907 3.5E-03 1.1E-02
Mixed NUD/AUD with DE/AX BIOTYPE
1 rs1030623 2 THSD7B 0.106 −2.080 3.1E-04 4.2E-01
2 rs998288 2 THSD7B 0.442 1.172 1.1E-03 9.8E-01
3 rs974529 2 THSD7B 0.191 1.455 2.3E-03 4.8E-01
Mixed AUD/NUD BIOTYPE
1 rs11646987*+ 16 TNRC6A 0.266 −0.566 7.3E-05 4.2E-02
2 rs990340 2 THSD7B 0.448 0.493 1.2E-04 7.7E-01
3 rs6497759 16 TNRC6A 0.196 −0.580 4.4E-04 4.3E-02
Open in a new tab

AUD: alcohol use disorder, NUD: nicotine use disorder, DE: depression, AX: anxiety, SNP: single nucleotide polymorphisms, Chr: chromosome, MAF: minor allele frequency.

*

P < 0.05 on training samples in the discovery set (n=594), Linkage-disequilibrium-adjusted and Bonferroni corrected. Wald test.

+

P < 0.05 on samples in the replication set (n=71). Wald test.

DISCUSSION

We biotyped individuals with AUD and NUD using connectivity markers. Three biotypes – primary NUD, mixed NUD/AUD with DE/AX, and mixed AUD/NUD, each with a distinct pattern of FCs and clinical characteristics – were differentiated from controls. Many of the FC features in support of the biotypes involved the cerebellum and visual cortex. Importantly, both biotype differentiation and MTL improved the accuracy of ANN in differentiating AUD and NUD subjects from HCs as evidenced by the replication results. The current findings highlighted the importance of biotyping of AUD and NUD subjects jointly to improve the performance of deep learning models in predicting AUD/NUD. Biotyping also identified a genetic marker that was not evident in the undifferentiated sample. The findings add to the effort to develop diagnostic biomarkers of AUD (59–61) and NUD (62,63) by reducing the heterogeneity in FC features to differentiate AUD and NUD subjects from controls, and increase the accuracy of classification by applying knowledge cross-learned from AUD (or NUD) samples in MTL.

FC features show highly non-linear relationships with psychiatric diagnoses (64,65). Linear dimension reduction (e.g., using principal component analysis) yielded poorer clustering solutions (see Supplementary Figure S4). As a non-linear algorithm, UMAP reduces the dimension of FC features via an iterative process that searches for coordinates in a lower dimensional space while preserving sample proximity. Here, the dimension reduction helped the K-means clustering of AUD/NUD subjects. The FC features contribute non-linearly to the classification; thus, we used Wilcoxon rank-sum tests to identify connectivity features that distinguish the biotypes.

The Primary NUD biotype exhibited altered insula and supramarginal gyrus (SMG) connectivity with the inferior frontal gyrus (pars orbitalis), relative to controls, consistent with previous findings of a key role of the insula in the maintenance of smoking addiction (66) and of reduced resting-state global efficiency of the SMG in heavy smokers in (67) The replicated FCs that differentiated the other two biotypes from HCs were of the cerebellum and secondary visual cortex (SVC). While the cerebellum is best known for its roles in conditioning, learning, and sensorimotor control, recent studies have highlighted its role in reward, motivation and cognitive control and its circuit dysfunction as a contributor to substance, including alcohol and nicotine, misuse (68–72). Earlier studies reported that alcohol dependence may influence resting-state FCs of the SVC (73,74) and that nicotine misuse impacts visual information processing (75). For instance, many have directly demonstrated changes in FCs of the visual cortex, amongst other brain regions, in chronic heavy drinking (76–78). Chronic nicotine exposure was associated with altered electroretinography and visual evoked potentials (79) as well as higher extra-striate regional activity in the absence of visual stimulation (80). Further, smoking induced a myriad of molecular changes in the retina that would impact visual cortical processing (81,82). Finally, characterized by higher depression and anxiety scores, visual cortex -related FCs identified with the mixed NUD/AUD with DE/AX showed more significant differences from HCs than those with the mixed AUD/NUD, which may reflect an association between visual impairment and depression (83,84).

The three biotypes could be distinguished, pairwise, with FCs that involved the cerebellum. Mixed NUD/AUD with DE/AX differed from both Mixed AUD/NUD and Primary NUD in cerebellar FC with the SVC, and Primary NUD and Mixed AUD/NUD differed in cerebellar FC with the primary sensory cortex. These findings are consistent with the roles of the cerebellum in affective functioning (85) and of cerebellar dysfunction in individuals engaged in alcohol misuse (86). With its connectivity with primary and association sensory cortices, the cerebellum is instrumentally involved in processing sensory information and cerebellar dysfunction may contribute to a multitude of perceptual disturbances noted in neuropsychiatric disorders (87).

A SNP (rs11646987) in TNRC6A was significantly associated with the mixed AUD/NUD biotype at an LD-adjusted Bonferroni corrected P-value of 7.3E-05 and was validated in the replication set (P = 0.04). TNRC6A was reported in an early GWAS to be associated with measures of alcohol (88,89) and nicotine misuse (90,91), consistent with the current findings. Notably, only with biotyping were we able to identify this SNP association with the AUD and NUD traits.

Our study has limitations. First, the discovery and replication samples were recruited from a single study; an independent test sample is needed to eliminate potential study-specific confounds. To maximize objectivity in validating the findings, we excluded the replication set from classifier training, in contrast to prior AUD and NUD classification studies (e.g., (92–97)). Second, despite the large size of the UKB data, only a small subset of subjects with concurrent rsfMRI, T1 anatomical and genetic data was available after quality control and exclusion. Thus, rather than performing a GWAS, we extracted SNPs from prior GWAS and identified one significantly associated SNP. Third, the UKB did not contain clinically-verified depression or anxiety or substance use metrics, and self-reported, low-resolution depression and anxiety scores or AUD/NUD codes could have limited the analysis. Further, the diagnostic codes used in this study came from patient health records gathered before imaging scans. No data were available on current drug use which likewise may impact the imaging findings. Fourth, the present study is cross-sectional, and it remains to be seen whether the identified connectivity features remain stable in biotyping smokers and drinkers. Finally, although we did not include other neural markers such as gray matter volumes and graph-theoretic network measures in our biotyping and classification scheme (98–101), future efforts should take advantage of the potential for use of multimodal imaging markers in biotyping substance use disorders.

Supplementary Material

1
2

ACKNOWLEDGEMENTS

This research has been conducted using the UK Biobank Resource under the Application 51296 – Machine Learning Analysis of Brain Images, Behaviors, and Genotypes to Understand Mental Disorders. The authors thank Dr. R. Todd Constable, Dr. Xilin Shen, and Corey Horien from the Yale Magnetic Resonance Research Center and Department of Diagnostic Radiology, Yale School of Medicine, who assisted Dr. Wuyi Wang in the early stage of fMRI data processing. The authors also thank Xinyu Wang from the Department of Computer Science and Engineering in the University of Connecticut who downloaded and processed the data from the UK Biobank website and identified clinical metrics for the study subjects.

FUNDING

The work was supported by an NIH grant 1R01 DA051922 to Drs. Bi and Li.

COMPETING INTERESTS

Dr. Kranzler is a member of advisory boards for Dicerna Pharmaceuticals, Sophrosyne Pharmaceuticals, and Enthion Pharmaceuticals; a consultant to Sobrera Pharmaceuticals; the recipient of research funding and medication supplies for an investigator-initiated study from Alkermes; a member of the American Society of Clinical Psychopharmacology’s Alcohol Clinical Trials Initiative, which was supported in the last three years by Alkermes, Dicerna, Ethypharm, Lundbeck, Mitsubishi, Otsuka, and Pear Therapeutics; and a holder of U.S. patent 10,900,082 titled: “Genotype-guided dosing of opioid agonists,” issued 26 January 2021. The other authors report no biomedical financial interests or potential conflicts of interest.

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

REFERENCES

  • 1.Pujades-Rodriguez M, George J, Shah AD, Rapsomaniki E, Denaxas S, West R, et al. (2015): Heterogeneous associations between smoking and a wide range of initial presentations of cardiovascular disease in 1 937 360 people in England: lifetime risks and implications for risk prediction. Int J Epidemiol 44: 129–141. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Litten RZ, Ryan ML, Falk DE, Reilly M, Fertig JB, Koob GF (2015): Heterogeneity of alcohol use disorder: understanding mechanisms to advance personalized treatment. Alcohol Clin Exp Res 39: 579–584. [DOI] [PubMed] [Google Scholar]
  • 3.Project MATCH Research Group (1998): Matching alcoholism treatments to client heterogeneity: Project MATCH three-year drinking outcomes. Alcohol Clin Exp Res 22: 1300–1311. [DOI] [PubMed] [Google Scholar]
  • 4.Babor TF, Del Boca FK (2003): Treatment Matching in Alcoholism. Cambridge University Press. [Google Scholar]
  • 5.Basu D, Ball SA, Feinn R, Gelernter J, Kranzler HR (2004): Typologies of drug dependence: comparative validity of a multivariate and four univariate models. Drug Alcohol Depend 73: 289–300. [DOI] [PubMed] [Google Scholar]
  • 6.Bierut LJ (2011): Genetic vulnerability and susceptibility to substance dependence. Neuron 69: 618–627. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Cuthbert BN, Insel TR (2013): Toward the future of psychiatric diagnosis: the seven pillars of RDoC. BMC Med 11: 1–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Cuthbert BN (2014): The RDoC framework: facilitating transition from ICD/DSM to dimensional approaches that integrate neuroscience and psychopathology. World psychiatry 13: 28–35. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Maj M (2016): Narrowing the gap between ICD/DSM and RDoC constructs: possible steps and caveats. World Psychiatry 15: 193. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Zhu T, Becquey C, Chen Y, Lejuez CW, Li C-SR, Bi J (2022): Identifying alcohol misuse biotypes from neural connectivity markers and concurrent genetic associations. Transl Psychiatry 12: 1–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Doyon WM, Thomas AM, Ostroumov A, Dong Y, Dani JA (2013): Potential substrates for nicotine and alcohol interactions: a focus on the mesocorticolimbic dopamine system. Biochem Pharmacol 86: 1181–1193. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.King A, Vena A, de Wit H, Grant JE, Cao D (2022): Effect of Combination Treatment With Varenicline and Nicotine Patch on Smoking Cessation Among Smokers Who Drink Heavily: A Randomized Clinical Trial. JAMA Netw Open 5: e220951–e220951. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Kimmey BA, Wittenberg RE, Croicu A, Shadani N, Ostroumov A, Dani JA (2022): The serotonin 2A receptor agonist TCB-2 attenuates heavy alcohol drinking and alcohol-induced midbrain inhibitory plasticity. Addiction Biology 27: e13147. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Carroll KM (2021): The profound heterogeneity of substance use disorders: Implications for treatment development. Curr Dir Psychol Sci 30: 358–364. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Wong CCY, Schumann G (2008): Genetics of addictions: strategies for addressing heterogeneity and polygenicity of substance use disorders. Philosophical Transactions of the Royal Society B: Biological Sciences 363: 3213–3222. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Buch AM, Liston C (2021): Dissecting diagnostic heterogeneity in depression by integrating neuroimaging and genetics. Neuropsychopharmacology 46: 156–175. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Sutherland MT, McHugh MJ, Pariyadath V, Stein EA (2012): Resting state functional connectivity in addiction: Lessons learned and a road ahead. Neuroimage 62: 2281–2295. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Camchong J, Stenger A, Fein G (2013): Resting-state synchrony in long-term abstinent alcoholics. Alcohol Clin Exp Res 37: 75–85. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Chanraud S, Pitel A-L, Pfefferbaum A, Sullivan EV (2011): Disruption of functional connectivity of the default-mode network in alcoholism. Cerebral cortex 21: 2272–2281. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Jasinska AJ, Zorick T, Brody AL, Stein EA (2014): Dual role of nicotine in addiction and cognition: a review of neuroimaging studies in humans. Neuropharmacology 84: 111–122. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.van Skike CE, Goodlett C, Matthews DB (2019): Acute alcohol and cognition: Remembering what it causes us to forget. Alcohol 79: 105–125. [DOI] [PubMed] [Google Scholar]
  • 22.Pitel AL, Segobin SH, Ritz L, Eustache F, Beaunieux H (2015): Thalamic abnormalities are a cardinal feature of alcohol-related brain dysfunction. Neurosci Biobehav Rev 54: 38–45. [DOI] [PubMed] [Google Scholar]
  • 23.Mooney-Leber SM, Gould TJ (2018): The long-term cognitive consequences of adolescent exposure to recreational drugs of abuse. Learning & Memory 25: 481–491. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Harvey RE, Berkowitz LE, Hamilton DA, Clark BJ (2019): The effects of developmental alcohol exposure on the neurobiology of spatial processing. Neurosci Biobehav Rev 107: 775–794. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Zahr NM, Pfefferbaum A (2017): Alcohol’s effects on the brain: neuroimaging results in humans and animal models. Alcohol Res. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Záborszky L, Gombkoto P, Varsanyi P, Gielow MR, Poe G, Role LW, et al. (2018): Specific basal forebrain–cortical cholinergic circuits coordinate cognitive operations. Journal of Neuroscience 38: 9446–9458. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Hu S, Ide JS, Chao HH, Zhornitsky S, Fischer KA, Wang W, et al. (2018): Resting state functional connectivity of the amygdala and problem drinking in non-dependent alcohol drinkers. Drug Alcohol Depend 185: 173–180. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Zhornitsky S, Ide JS, Wang W, Chao HH, Zhang S, Hu S, et al. (2018): Problem drinking, alcohol expectancy, and thalamic resting-state functional connectivity in nondependent adult drinkers. Brain Connect 8: 487–502. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Chen Y, Dhingra I, Chaudhary S, Fucito L, Li C-SR (2022): Overnight abstinence is associated with smaller secondary somatosensory cortical volumes and higher somatosensory-motor cortical functional connectivity in cigarette smokers. Nicotine and Tobacco Research 24: 1889–1897. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Zhang S, Hu S, Fucito LM, Luo X, Mazure CM, Zaborszky L, Li CR (2017): Resting-state functional connectivity of the basal nucleus of meynert in cigarette smokers: dependence level and gender differences. Nicotine & Tobacco Research 19: 452–459. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Arienzo D, Happer JP, Molnar SM, Alderson-Myers A, Marinkovic K (2020): Binge drinking is associated with altered resting state functional connectivity of reward-salience and top down control networks. Brain Imaging Behav 14: 1731–1746. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Galandra C, Basso G, Manera M, Crespi C, Giorgi I, Vittadini G, et al. (2019): Abnormal fronto-striatal intrinsic connectivity reflects executive dysfunction in alcohol use disorders. Cortex 115: 27–42. [DOI] [PubMed] [Google Scholar]
  • 33.Gerchen MF, Rentsch A, Kirsch M, Kiefer F, Kirsch P (2019): Shifts in the functional topography of frontal cortex-striatum connectivity in alcohol use disorder. Addiction Biology 24: 1245–1253. [DOI] [PubMed] [Google Scholar]
  • 34.Heitzeg MM, Nigg JT, Yau W-YW, Zucker RA, Zubieta J-K (2010): Striatal dysfunction marks preexisting risk and medial prefrontal dysfunction is related to problem drinking in children of alcoholics. Biol Psychiatry 68: 287–295. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Heitzeg MM, Nigg JT, Yau W-YW, Zubieta J-K, Zucker RA (2008): Affective circuitry and risk for alcoholism in late adolescence: differences in frontostriatal responses between vulnerable and resilient children of alcoholic parents. Alcohol Clin Exp Res 32: 414–426. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Kohno M, Dennis LE, McCready H, Hoffman WF (2017): Executive control and striatal resting-state network interact with risk factors to influence treatment outcomes in alcohol-use disorder. Front Psychiatry 8: 182. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Rose EJ, Ross TJ, Salmeron BJ, Lee M, Shakleya DM, Huestis MA, Stein EA (2013): Acute nicotine differentially impacts anticipatory valence-and magnitude-related striatal activity. Biol Psychiatry 73: 280–288. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Sweitzer MM, Geier CF, Joel DL, McGurrin P, Denlinger RL, Forbes EE, Donny EC (2014): Dissociated effects of anticipating smoking versus monetary reward in the caudate as a function of smoking abstinence. Biol Psychiatry 76: 681–688. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Froeliger B, McConnell PA, Stankeviciute N, McClure EA, Kalivas PW, Gray KM (2015): The effects of N-acetylcysteine on frontostriatal resting-state functional connectivity, withdrawal symptoms and smoking abstinence: a double-blind, placebo-controlled fMRI pilot study. Drug Alcohol Depend 156: 234–242. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Sudlow C, Gallacher J, Allen N, Beral V, Burton P, Danesh J, et al. (2015): UK biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLoS Med 12: e1001779. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.World Health Organization (1993): The ICD-10 Classification of Mental and Behavioural Disorders. World Health Organization. [Google Scholar]
  • 42.Miller KL, Alfaro-Almagro F, Bangerter NK, Thomas DL, Yacoub E, Xu J, et al. (2016): Multimodal population brain imaging in the UK Biobank prospective epidemiological study. Nat Neurosci 19: 1523–1536. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Alfaro-Almagro F, Jenkinson M, Bangerter NK, Andersson JLR, Griffanti L, Douaud G, et al. (2018): Image processing and Quality Control for the first 10,000 brain imaging datasets from UK Biobank. Neuroimage 166: 400–424. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Ashburner J, Barnes G, Chen C-C, Daunizeau J, Flandin G, Friston K, et al. (2014): SPM12 manual. Wellcome Trust Centre for Neuroimaging, London, UK: 2464: 4. [Google Scholar]
  • 45.Satterthwaite TD, Elliott MA, Gerraty RT, Ruparel K, Loughead J, Calkins ME, et al. (2013): An improved framework for confound regression and filtering for control of motion artifact in the preprocessing of resting-state functional connectivity data. Neuroimage 64: 240–256. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Shen X, Tokoglu F, Papademetris X, Constable RT (2013): Groupwise whole-brain parcellation from resting-state fMRI data for network node identification. Neuroimage 82: 403–415. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Bycroft C, Freeman C, Petkova D, Band G, Elliott LT, Sharp K, et al. (2017): Genome-wide genetic data on~ 500,000 UK Biobank participants. BioRxiv 166298. [Google Scholar]
  • 48.Buniello A, MacArthur JAL, Cerezo M, Harris LW, Hayhurst J, Malangone C, et al. (2019): The NHGRI-EBI GWAS Catalog of published genome-wide association studies, targeted arrays and summary statistics 2019. Nucleic Acids Res 47: D1005--D1012. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 49.Padula CB, Anthenelli RM, Eliassen JC, Nelson E, Lisdahl KM (2015): Gender effects in alcohol dependence: an fMRI pilot study examining affective processing. Alcohol Clin Exp Res 39: 272–281. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 50.Pfefferbaum A, Rosenbloom M, Crusan K, Jernigan TL (1988): Brain CT changes in alcoholics: effects of age and alcohol consumption. Alcohol Clin Exp Res 12: 81–87. [DOI] [PubMed] [Google Scholar]
  • 51.McInnes L, Healy J, Melville J (2018): Umap: Uniform manifold approximation and projection for dimension reduction. arXiv preprint arXiv:180203426. [Google Scholar]
  • 52.Shutaywi M, Kachouie NN (2021): Silhouette analysis for performance evaluation in machine learning with applications to clustering. Entropy 23: 759. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 53.Zhao S, Sun J, Shimizu K, Kadota K (2018): Silhouette scores for arbitrary defined groups in gene expression data and insights into differential expression results. Biol Proced Online 20: 1–12. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 54.Zhu J, Qiu A (2022): Chinese adult brain atlas with functional and white matter parcellation. Sci Data 9: 1–10. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 55.Johnson RW (2001): An introduction to the bootstrap. Teach Stat 23: 49–54. [Google Scholar]
  • 56.Zhou X, Stephens M (2012): Genome-wide efficient mixed-model analysis for association studies. Nat Genet 44: 821. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 57.Duggal P, Gillanders EM, Holmes TN, Bailey-Wilson JE (2008): Establishing an adjusted p-value threshold to control the family-wide type 1 error in genome wide association studies. BMC Genomics 9: 1–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 58.Xia M, Wang J, He Y (2013): BrainNet Viewer: a network visualization tool for human brain connectomics. PLoS One 8: e68910. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 59.Windle M, Scheldt DM (2004): Alcoholic subtypes: Are two sufficient? Addiction 99: 1508–1519. [DOI] [PubMed] [Google Scholar]
  • 60.Müller M, Ajdacic-Gross V, Vetrella AB, Preisig M, Castelao E, Lasserre A, et al. (2020): Subtypes of alcohol use disorder in the general population: A latent class analysis. Psychiatry Res 285: 112712. [DOI] [PubMed] [Google Scholar]
  • 61.Schuckit MA, Smith TL, Paulus MP, Tapert SF, Simmons AN, Tolentino NJ, Shafir A (2016): The ability of functional magnetic resonance imaging to predict heavy drinking and alcohol problems 5 years later. Alcohol Clin Exp Res 40: 206–213. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 62.Manley MJ, de Jonge P, Kershaw TS, Desai RA, Lin H, Kasl S v (2009): Association of major depression with subtypes of nicotine dependence found among adult daily smokers: a latent class analysis. Drug Alcohol Depend 104: 126–132. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 63.Meyer C, Rumpf H-J, Schumann A, Hapke U, John U (2004): Subtyping general population smokers not intending to quit by stages to reduce smoking. Nicotine & tobacco research 6: 1043–1050. [DOI] [PubMed] [Google Scholar]
  • 64.Hlinka J, Paluš M, Vejmelka M, Mantini D, Corbetta M (2011): Functional connectivity in resting-state fMRI: is linear correlation sufficient? Neuroimage 54: 2218–2225. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 65.Laird AR, Rogers BP, Meyerand ME (2002): Investigating the nonlinearity of fMRI activation data. Proceedings of the Second Joint 24th Annual Conference and the Annual Fall Meeting of the Biomedical Engineering Society, vol. 11: 11–12. [Google Scholar]
  • 66.Bi Y, Yuan K, Guan Y, Cheng J, Zhang Y, Li Y, et al. (2017): Altered resting state functional connectivity of anterior insula in young smokers. Brain Imaging Behav 11: 155–165. [DOI] [PubMed] [Google Scholar]
  • 67.Fedota JR, Stein EA (2015): Resting-state functional connectivity and nicotine addiction: prospects for biomarker development. Ann N Y Acad Sci 1349: 64–82. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 68.Moulton EA, Elman I, Becerra LR, Goldstein RZ, Borsook D (2014): The cerebellum and addiction: insights gained from neuroimaging research. Addiction biology 19: 317–331. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 69.Sullivan EV, Rosenbloom MJ, Deshmukh Anjali JE, Pfefferbaum A (1995): Alcohol and the Cerebellum: Effects on Balance, Motor Coordination, and Cognition. Alcohol Health Res World 19: 138–141. [PMC free article] [PubMed] [Google Scholar]
  • 70.Shanmugarajah PD, Hoggard N, Currie S, Aeschlimann DP, Aeschlimann PC, Gleeson DC, et al. (2016): Alcohol-related cerebellar degeneration: not all down to toxicity? Cerebellum Ataxias 3: 1–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 71.Kekkonen V, Kallioniemi E, Kaarre O, Könönen M, Kivimäki P, Gröhn H, et al. (2021): Heavy drinking from adolescence to young adulthood is associated with an altered cerebellum. Alcohol 92: 35–40. [DOI] [PubMed] [Google Scholar]
  • 72.Sutherland MT, Riedel MC, Flannery JS, Yanes JA, Fox PT, Stein EA, Laird AR (2016): Chronic cigarette smoking is linked with structural alterations in brain regions showing acute nicotinic drug-induced functional modulations. Behavioral and Brain Functions 12: 1–15. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 73.Luo X, Guo L, Dai X-J, Wang Q, Zhu W, Miao X, Gong H (2017): Abnormal intrinsic functional hubs in alcohol dependence: evidence from a voxelwise degree centrality analysis. Neuropsychiatr Dis Treat 13: 2011. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 74.Fein G, Shimotsu R, Chu R, Barakos J (2009): Parietal gray matter volume loss is related to spatial processing deficits in long-term abstinent alcoholic men. Alcohol Clin Exp Res 33: 1806–1814. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 75.Lawrence NS, Ross TJ, Stein EA (2002): Cognitive mechanisms of nicotine on visual attention. Neuron 36: 539–548. [DOI] [PubMed] [Google Scholar]
  • 76.Ochi R, Ueno F, Sakuma M, Tani H, Tsugawa S, Graff-Guerrero A, et al. (2022): Patterns of functional connectivity alterations induced by alcohol reflect somatostatin interneuron expression in the human cerebral cortex. Sci Rep 12: 7896. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 77.Shokri-Kojori E, Tomasi D, Wiers CE, Wang G-J, Volkow ND (2017): Alcohol affects brain functional connectivity and its coupling with behavior: greater effects in male heavy drinkers. Mol Psychiatry 22: 1185–1195. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 78.Vergara VM, Liu J, Claus ED, Hutchison K, Calhoun V (2017): Alterations of resting state functional network connectivity in the brain of nicotine and alcohol users. Neuroimage 151: 45–54. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 79.Naser NT, Zemon VM, Varghese SB, Keyser KT, Hartmann EE (2012): Effects of nicotine on processing in the visual pathways. Invest Ophthalmol Vis Sci 53: 5735. [Google Scholar]
  • 80.Tanabe J, Nyberg E, Martin LF, Martin J, Cordes D, Kronberg E, Tregellas JR (2011): Nicotine effects on default mode network during resting state. Psychopharmacology (Berl) 216: 287–295. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 81.Gonçalves RGJ, Vasques JF, Trindade P, Serfaty CA, Campello-Costa P, Faria-Melibeu AC (2016): Nicotine-induced plasticity in the retinocollicular pathway: Evidence for involvement of amyloid precursor protein. Neuroscience 313: 1–9. [DOI] [PubMed] [Google Scholar]
  • 82.Wang V, Heffer A, Roztocil E, Feldon SE, Libby RT, Woeller CF, Kuriyan AE (2022): TNF-α and NF-κB signaling play a critical role in cigarette smoke-induced epithelial-mesenchymal transition of retinal pigment epithelial cells in proliferative vitreoretinopathy. PLoS One 17: e0271950. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 83.Tsai S-Y, Cheng C-Y, Hsu W-M, Su TP, Liu J-H, Chou P (2003): Association between visual impairment and depression in the elderly. Journal of the Formosan Medical Association 102: 86–90. [PubMed] [Google Scholar]
  • 84.Xia M, Si T, Sun X, Ma Q, Liu B, Wang L, et al. (2019): Reproducibility of functional brain alterations in major depressive disorder: Evidence from a multisite resting-state functional MRI study with 1,434 individuals. Neuroimage 189: 700–714. [DOI] [PubMed] [Google Scholar]
  • 85.Turner BM, Paradiso S, Marvel CL, Pierson R, Ponto LLB, Hichwa RD, Robinson RG (2007): The cerebellum and emotional experience. Neuropsychologia 45: 1331–1341. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 86.Luo J (2015): Effects of ethanol on the cerebellum: advances and prospects. The Cerebellum 14: 383–385. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 87.Baumann O, Borra RJ, Bower JM, Cullen KE, Habas C, Ivry RB, et al. (2015): Consensus paper: the role of the cerebellum in perceptual processes. The Cerebellum 14: 197–220. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 88.Brazel DM, Jiang Y, Hughey JM, Turcot V, Zhan X, Gong J, et al. (2019): Exome chip meta-analysis fine maps causal variants and elucidates the genetic architecture of rare coding variants in smoking and alcohol use. Biol Psychiatry 85: 946–955. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 89.Zhou H, Sealock JM, Sanchez-Roige S, Clarke T-K, Levey DF, Cheng Z, et al. (2020): Genome-wide meta-analysis of problematic alcohol use in 435,563 individuals yields insights into biology and relationships with other traits. Nat Neurosci 23: 809–818. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 90.Pasman JA, Demange PA, Guloksuz S, Willemsen AHM, Abdellaoui A, ten Have M, et al. (2022): Genetic Risk for Smoking: Disentangling Interplay Between Genes and Socioeconomic Status. Behav Genet 52: 92–107. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 91.Liu M, Jiang Y, Wedow R, Li Y, Brazel DM, Chen F, et al. (2019): Association studies of up to 1.2 million individuals yield new insights into the genetic etiology of tobacco and alcohol use. Nat Genet 51: 237–244. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 92.Zhu X, Du X, Kerich M, Lohoff FW, Momenan R (2018): Random forest based classification of alcohol dependence patients and healthy controls using resting state MRI. Neurosci Lett 676: 27–33. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 93.Kamarajan C, Ardekani BA, Pandey AK, Kinreich S, Pandey G, Chorlian DB, et al. (2020): Random Forest Classification of Alcohol Use Disorder Using fMRI Functional Connectivity, Neuropsychological Functioning, and Impulsivity Measures. Brain Sci 10. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 94.Kamarajan C, Ardekani BA, Pandey AK, Chorlian DB, Kinreich S, Pandey G, et al. (2020): Random forest classification of alcohol use disorder using EEG source functional connectivity, neuropsychological functioning, and impulsivity measures. Behavioral Sciences 10: 62. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 95.Pariyadath V, Stein EA, Ross TJ (2014): Machine learning classification of resting state functional connectivity predicts smoking status. Front Hum Neurosci 8: 425. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 96.Wang S, Zhang R, Deng Y, Chen K, Xiao D, Peng P, Jiang T (2018): Discrimination of smoking status by MRI based on deep learning method. Quant Imaging Med Surg 8: 1113. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 97.Tahmassebi A, Gandomi AH, Schulte MHJ, Goudriaan AE, Foo SY, Meyer-Baese A (2018): Optimized naive-bayes and decision tree approaches for fMRI smoking cessation classification. Complexity 2018. [Google Scholar]
  • 98.Mackey S, Allgaier N, Chaarani B, Spechler P, Orr C, Bunn J, et al. (2019): Mega-analysis of gray matter volume in substance dependence: general and substance-specific regional effects. American Journal of Psychiatry 176: 119–128. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 99.Guggenmos M, Schmack K, Veer IM, Lett T, Sekutowicz M, Sebold M, et al. (2020): A multimodal neuroimaging classifier for alcohol dependence. Sci Rep 10: 1–12. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 100.Tomasi D, Wiers CE, Manza P, Shokri-Kojori E, Michele-Vera Y, Zhang R, et al. (2021): Accelerated aging of the amygdala in alcohol use disorders: relevance to the dark side of addiction. Cerebral Cortex 31: 3254–3265. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 101.Holla B, Biswal J, Ramesh V, Shivakumar V, Bharath RD, Benegal V, et al. (2020): Effect of prefrontal tDCS on resting brain fMRI graph measures in Alcohol Use Disorders: A randomized, double-blind, sham-controlled study. Prog Neuropsychopharmacol Biol Psychiatry 102: 109950. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1
NIHMS1947421-supplement-1.pdf (1.8MB, pdf)
2
NIHMS1947421-supplement-2.xlsx (112.4KB, xlsx)

RESOURCES